Handbook of Computational Intelligence in Biomedical Engineering and Healthcare

By Asit Kumar Das

Handbook of Computational Intelligence in Biomedical Engineering and Healthcare helps readers analyze and conduct advanced research in specialty healthcare applications surrounding oncology, genomics and genetic data, ontologies construction, bio-memetic systems, biomedical electronics, protein structure prediction, and biomedical data analysis. The book provides the reader with a comprehensive guide to advanced computational intelligence, spanning deep learning, fuzzy logic, connectionist systems, evolutionary computation, cellular automata, self-organizing systems, soft computing, and hybrid intelligent systems in biomedical and healthcare applications. Sections focus on important biomedical engineering applications, including biosensors, enzyme immobilization techniques, immuno-assays, and nanomaterials for biosensors and other biomedical techniques.

Other sections cover gene-based solutions and applications through computational intelligence techniques and the impact of nonlinear/unstructured data on experimental analysis.

• Presents a comprehensive handbook that covers an Introduction to Computational Intelligence in Biomedical Engineering and Healthcare, Computational Intelligence Techniques, and Advanced and Emerging Techniques in Computational Intelligence
• Helps readers analyze and do advanced research in specialty healthcare applications
• Includes links to websites, videos, articles and other online content to expand and support primary learning objectives

• Language: English
• Publisher: Academic Press
• Release date: Apr 8, 2021
• ISBN: 9780128222614

Book preview

Handbook of Computational Intelligence in Biomedical Engineering and Healthcare

Editors

Janmenjoy Nayak

Bighnaraj Naik

Danilo Pelusi

Asit Kumar Das

Table of Contents

Cover image

Title page

Copyright

Contributors

Biographies

Preface

Chapter 1. Application of dynamical systems based deep learning algorithms to model emergent characteristics for healthcare diagnostics

1. Introduction

2. Deep learning applications for brainwaves monitoring

3. Healthcare Modeling and simulation using feedback hybrid artificial neural networks

4. Derivative estimation using feedback networks

5. Usage of deep learning knowledge mining in Hybrid Inference Networks

6. Conclusions

Chapter 2. Computational intelligence in healthcare and biosignal processing

1. Introduction

2. Investigation on various deep clustering algorithms

3. Investigation on clustering algorithms for the unsupervised learning methodology

4. Conclusion

Chapter 3. A semi-supervised approach for automatic detection and segmentation of optic disc from retinal fundus image

1. Introduction

2. State-of-the-art

3. Proposed method

4. Experimentations and results

5. Conclusions

Chapter 4. Medical decision support system using data mining: an intelligent health care monitoring system for guarded travel

1. Introduction

2. Related works

3. Proposed system

4. Performance analysis

5. Conclusion

Chapter 5. Deep learning in gastroenterology: a brief review

1. Introduction

2. Anomalies in GI-tract and medical image modalities for GE

3. Conventional-ML in gastroenterology

4. DL based GI-tract diagnosis system

5. Critical analysis and discussions

6. Conclusion

Chapter 6. Application of soft computing techniques to calculation of medicine dose during the treatment of patient: a fuzzy logic approach

1. Introduction

2. Soft computing

3. Fuzzy logic

4. Fuzzy logic based intelligent system

5. Comparison of drug doses suggested by expert doctor and proposed fuzzy based intelligent system

6. Conclusion

Chapter 7. Multiobjective optimization technique for gene selection and sample categorization

1. Introduction

2. Gene subset selection

3. Results and discussions

4. Conclusion and future work

Chapter 8. Medical decision support system using data mining semicircular-based angle-oriented facial recognition using neutrosophic logic

1. Introduction

2. Semicircular model based angle oriented images

3. Angle-oriented fuzzy rough sets

4. Ternary relationship with angle-oriented face recognition

5. K-means fuzzy rough angle-oriented clusters

6. Neutrosophic logic

7. Hyperplane

8. Evolutionary optimization method

9. Rotation and reduction procedure (R2 procedure)

10. Experimental result

11. Conclusion

Chapter 9. Preservation module prediction by weighted differentially coexpressed gene network analysis (WDCGNA) of HIV-1 disease: a case study for cancer

1. Introduction

2. Related work

3. Material and methods

4. Result and analysis

5. KEGG pathway analysis

6. Conclusion

Chapter 10. Computational intelligence for genomic data: a network biology approach

1. Introduction

2. Next generation sequencing overview

3. Different sequencing platforms

4. Different scores and parameters involved in biological network

5. Genomic data mining and biological network analysis: a case study

6. Summary and conclusions

Chapter 11. A Kinect-based motor rehabilitation system for stroke recovery

1. Introduction

2. Literature survey

3. Proposed work

4. Experimental results

5. Conclusion and future work

Chapter 12. Empirical study on Uddanam chronic kidney diseases (UCKD) with statistical and machine learning analysis including probabilistic neural networks

1. Introduction

2. Literature survey

3. Proposal model and materials

4. Results and discussions

5. Conclusion and social benefits

Chapter 13. Enhanced brain tumor detection using fractional wavelet transform and artificial neural network

1. Introduction

2. Literature survey

3. Fractional wavelet transform

4. Principal component analysis

5. Artificial neural network

6. Proposed method

7. Experimental results

8. Conclusion

Chapter 14. A study on smartphone sensor-based Human Activity Recognition using deep learning approaches

1. Introduction

2. Literature survey

3. Dataset description

4. Architecture of different deep networks

5. Results and discussion

6. Conclusion and future work

Index

Copyright

Academic Press is an imprint of Elsevier

125 London Wall, London EC2Y 5AS, United Kingdom

525 B Street, Suite 1650, San Diego, CA 92101, United States

50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States

The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom

Copyright © 2021 Elsevier Inc. All rights reserved.

No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Details on how to seek permission, further information about the Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions.

This book and the individual contributions contained in it are protected under copyright by the Publisher (other than as may be noted herein).

Notices

Knowledge and best practice in this field are constantly changing. As new research and experience broaden our understanding, changes in research methods, professional practices, or medical treatment may become necessary.

Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, methods, compounds, or experiments described herein. In using such information or methods they should be mindful of their own safety and the safety of others, including parties for whom they have a professional responsibility.

To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the material herein.

Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of Congress

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library

ISBN: 978-0-12-822260-7

For information on all Academic Press publications visit our website at https://www.elsevier.com/books-and-journals

Publisher: Mara Conner

Acquisitions Editor: Chris Katsaropoulos

Editorial Project Manager: Emily Thomson

Production Project Manager: Selvaraj Raviraj

Cover Designer: Miles Hitchen

Typeset by TNQ Technologies

Contributors

David Al-Dabass,     School of Science and Technology, Nottingham Trent University, Nottingham, United Kingdom

Nagaraj Balakrishnan,     Department of Electronics and Communication Engineering, Rathinam Technical Campus, Coimbatore, Tamil Nadu, India

Valentina E. Balas,     Automation and Applied Informatics, Aurel Vlaicu University of Arad, Romania, Arad, Romania

Bandana Barman,     Department of Electronics & Communication Engineering, Kalyani Government Engineering College, Kalyani, West Bengal, India

Sayanwita Barua,     Department of Computer Science and Engineering, Jadavpur University, Kolkata, West Bengal, India

Debotosh Bhattacharjee,     Department of Computer Science and Engineering, Jadavpur University, Kolkata, West Bengal, India

Gajanan K. Birajdar,     Department of Electronics Engineering, Ramrao Adik Institute of Technology, Navi Mumbai, Maharashtra, India

Asit Kumar Das,     Department of Computer Science and Technology, IIEST, Howrah, West Bengal, India

Neha Das,     Department of Computer Science and Engineering, Maulana Abul Kalam Azad University of Technology, Kolkata, West Bengal, India

Sunanda Das,     Department of Computer Science and Engineering, SVCET, Chittoor, Andhra Pradesh, India

L. Jegatha Deborah,     Department of Computer Science and Engineering, University College of Engineering, Tindivanam, Tamil Nadu, India

Susovan Jana,     School of Education Technology, Jadavpur University, Kolkata, West Bengal, India

Ria Kanjilal,     Department of Electronics & Communication Engineering, Kalyani Government Engineering College, Kalyani, West Bengal, India

Bhakti Kaushal,     Department of Electronics and Telecommunication Engineering, Ramrao Adik Institute of Technology, Navi Mumbai, Maharashtra, India

Fahmida Khan,     Department of Chemistry, National Institute of Technology, Raipur, Chhattisgarh, India

Mainak Kumar Kundu,     Department of Electronics & Communication Engineering, Brainware Engineering College, Barasat, Kolkata, West Bengal, India

Raffaele Mascella,     Faculty of Communication Sciences, University of Teramo, Teramo, Italy

Lela Mirtskhulava,     Iv. Javakhishvili Tbilisi State University/San Diego State University, Tbilisi, Georgia

Manohar Mishra,     Department of Electrical and Electronics Engineering, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan (Deemed to be) University, Bhubaneswar, Odisha, India

R.N.V. Jagan Mohan,     Department of Computer Science and Engineering, Sagi Rama Krishnam Raju Engineering College, Bhimavaram, Andhra Pradesh, India

Subhashree Mohapatra,     Department of Computer Science and Engineering, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan (Deemed to be) University, Bhubaneswar, Odisha, India

Riktim Mondal,     Department of Computer Science and Engineering, Jadavpur University, Kolkata, West Bengal, India

Dibyendu Mukhopadhyay,     Department of Computer Science and Engineering, Jadavpur University, Kolkata, West Bengal, India

Ranjan Parekh,     School of Education Technology, Jadavpur University, Kolkata, West Bengal, India

Mukesh D. Patil,     Department of Electronics and Telecommunication Engineering, Ramrao Adik Institute of Technology, Navi Mumbai, Maharashtra, India

Subrat Kumar Pattanayak,     Department of Chemistry, National Institute of Technology, Raipur, Chhattisgarh, India

Arunkumar Rajendran,     Department of Electronics and Communication Engineering, Rathinam Technical Campus, Coimbatore, Tamil Nadu, India

S.C. Rajkumar,     Department of Computer Science and Engineering, University College of Engineering, Panruti, Tamil Nadu, India

Sriparna Saha,     Department of Computer Science and Engineering, Maulana Abul Kalam Azad University of Technology, Kolkata, West Bengal, India

Parameswar Sahu,     Indian Council of Agricultural Research (ICAR-CIFRI), Barrackpore, West Bengal, India

Bijan Sarkar,     Department of Production Engineering, Jadavpur University, Kolkata, West Bengal, India

Ram Sarkar,     Department of Computer Science and Engineering, Jadavpur University, Kolkata, West Bengal, India

Pawan Kumar Singh,     Department of Information Technology, Jadavpur University, Kolkata, West Bengal, India

Tripti Swarnkar,     Department of Computer Application, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan (Deemed to be) University, Bhubaneswar, Odisha, India

P. Vijayakumar,     Department of Computer Science and Engineering, University College of Engineering, Tindivanam, Tamil Nadu, India

T. PanduRanga Vital,     Department of Computer Science and Engineering, Aditya Institute of Technology and Management, Srikakulam, Andhra Pradesh, India

Ramjeet Singh Yadav,     Department of Computer Science and Engineering, Ashoka Institute of Technology and Management, Varanasi, Uttar Pradesh, India

Biographies

Janmenjoy Nayak is working as an associate professor at the Department of Computer Science and Engineering, Aditya Institute of Technology and Management (AITAM), Tekkali, K Kotturu, AP- 532201, India. He has published more than 110 research papers in various reputed peer reviewed referred journals, international conferences, and book chapters. He is the recipient of the best researcher award from Jawaharlal Nehru University of Technology, Kakinada, Andhra Pradesh for 2018–2019, and many other awards. His area of interest includes data mining, nature inspired algorithms and soft computing. He has edited nine books from various publishers such as Elsevier and Springer.

Bighnaraj Naik is an assistant professor in the Department of Computer Application, Veer SurendraSai University of Technology (Formerly UCE Burla), Odisha, India. He has published more than 90 research papers in various reputed peer reviewed international journals, conferences, and book chapters. He has edited eleven books from various publishers such as Elsevier, Springer, and IGI Global. At present, he has more than 10   years of teaching experience in the field of Computer Science and IT. He is a member of IEEE. His area of interest includes data mining, computational intelligence, soft computing, and its applications.

Danilo Pelusi is working as an associate professor at the Faculty of Communication Sciences, University of Teramo. As Associate Editor of IEEE Transactions on Emerging Topics in Computational Intelligence, IEEE Access, International Journal of Machine Learning and Cybernetics (Springer), and Array (Elsevier), he served as guest editor for Elsevier, Springer, and Inderscience Journals and as program member of many conferences, as well as editorial board member of many journals. Reviewer of reputed journals such as IEEE Transactions on Fuzzy Systems and IEEE Transactions on Neural Networks and Machine Learning, his research interests include fuzzy logic, neural networks, information theory, and evolutionary algorithms.

Asit Kumar Das is a professor of the Department of Computer Science and Technology, Indian Institute of Engineering Science and Technology Shibpur, Howrah, and currently acting as the head of the Center of Healthcare Science and Technology of his institute. He has published one research monograph, three edited books, many book chapters, and over 100 research articles in peer-reviewed journals and international conferences. His current research interests include data mining and pattern recognition, social network analysis, evolutionary computing, text, audio, and video processing. Prof. Das has already guided five PhD scholars, and seven more scholars are currently working under him.

Preface

In some coming decade, the expansion of medical data (both structured and unstructured) will present issues as well as prospects for hospitals and academics. The current situation of storage of amount of data is quite huge in contemporary databases due to the availability and popularity of the internet as well as cloud sharing. Healthcare has always been a challenging field and needs more focus. The advantages of intelligent computing have been extensively discussed in the medical literature. Computational intelligence includes the ability to use refined algorithms to learn correct features from a bulky healthcare data, which may be used to obtain insights to assist clinical practice. Automated systems can aid physicians by providing current medical information from news articles, journals, textbooks, and clinical practices to inform proper patient care. For the past decade, there is an aggressive increment in the development of computational intelligence methods. Data produced by the medical organizations is extremely huge and multifaceted due to which it is hard to examine the information in order to make significant conclusion regarding the health of a patient. This information holds particulars concerning hospitals, patients, medical claims, cost of treatments, etc. So, there is a demand to generate a commanding tool for examining and removing vital information from this difficult data. Thus the information needs to be reviewed and structured in order to preserve efficient decision-making. In the present scenario of computing, computational intelligence tools present adaptive mechanisms that permit the understanding of difficult data and altering environments. The results of computational intelligence technologies are to present profits to medical fields for assembling the patients having same type of diseases or fitness problems, so that medical organization gives them effectual treatments.

In spite of much advancement, we believe that machines (intelligent computing methods) cannot replace human physicians in the upcoming future, but it can definitely assist physicians to make better clinical decisions or even replace human judgment in certain functional areas of healthcare. Healthcare is a multifaceted domain, which includes advanced decision-making, remote monitoring, healthcare logistics, operational excellence, and modern information systems. This proposed book is more focused and intricate toward complex problem solving with integration of computational intelligence, biomedical, and healthcare applications. The book is a voluminous collection for the state of the art as well as advances of different computational intelligence methods and will provide effective solutions for the confront faced by the healthcare institutions and hospitals for effective analysis, storage and analysis of data. Various methods such as Dynamical Systems based on Deep Learning, Biosignal Processing, Decision Support System using Data-Mining, Fuzzy Logic, Multiobjective Optimization Technique, Neutrosophic Logic, Probabilistic Neural Networks, Fractional Wavelet Transform, etc. are discussed with several applications to meet the requirements of the latest biomedical and healthcare challenges.

Chapter 1 discusses the application of dynamical systems based on deep learning algorithms to model emergent behavior for healthcare diagnostics. The algorithms and methodologies based on dynamical-system deep learning are used to formulate models for healthcare variables. Further, hybrid recurrent nets are proposed to construct deep learning models of observed trajectory patterns of these variables. Each observed trajectory is subjected to a deep learning mining process to determine its dynamical behavior parameters. Results obtained from this study using the simulation demonstrated the quality of the algorithms in dealing with the range of difficulties inherent in the problem.

Chapter 2 focuses on the enhancement of the behavior and nature of the deep learning method with clustering algorithm. The major objective is to analyze the impacts of unsupervised algorithms in context to deep learning. Methods such as autoencoder with clustering, Deep Embedded Clustering, Discriminately Boosted Clustering, image clustering, Deep Embedded Regularized Clustering, Variation Deep Embedding, Information Maximizing Generative Adversarial Network, Joint unsupervised Learning, Deep Adaptive Image Clustering, and the Deep Clustering framework based on Orthogonal AutoEncoder are thoroughly discussed with their challenges and issues in processing medical image data.

Chapter 3 proposes an automatic optic disc detection and segmentation technique to address the task of optic disc in retina. With the converted grayscale image and noise removal steps, the authors have used edge detection operation to find the edges of the optic disc. The possible optic disc center from the edges is calculated using Circular Hough Transform technique. With a supervised machine learning algorithm, the actual and perfect optic disc region among the candidate regions for the optic disc has been computed.

Chapter 4 is about the development of an intelligent healthcare system to monitor the driver's heart rate that utilizes an artificial recurrent neural network (RNN) model, which continually monitors driver's health condition and persistently update to the cloud server. To accomplish a reliable emergency service, the proposed proximity based communication model transmits the critical condition record to the service providers even without internet capability. The Long-Short Term Memory (LSTM) is capable of learning the driver's behavior continually in long-term decencies of his/her activity which converts sensor reading into optimized EHR. The proposed work would be noticed as one step toward guaranteed guarded journey.

Chapter 5 discussed the application of AI as a solution to Gastroenterology-based medical diagnosis. In Gastroenterology-based research area, general practitioner handles huge amounts of medical data and numerous varieties of imaging instruments. The artificial intelligence has been successfully applied in the field of GE for effective diagnosis and analyzing gastrointestinal images. The chapter summarize the AI applications in Gastroenterology with several key issues/challenges. It briefly enlighten about the Endoscopy, X-ray, Ultrasound, CT-scan, MRI, PET, etc. using machine learning and deep learning based approaches.

Chapter 6 proposed a fuzzy logic based intelligent system calculation of medicine doses for chronic intestine illness symptoms like sedimentation and prostate antigen. As the medicine doses play a vital role during the procedure and observation of the patient, appropriate medicine dose require for the long-duration treatment of any patient. The chapter briefly discusses about the components of fuzzy membership functions, rules and inference system used for the calculation of medicine dose. With several experiments the authors claimed that, their proposed method is a suitable intelligent based framework for determining the dose of medicine given to patients with chronic intestinal infections.

Chapter 7 developed a multiobjective optimization technique based on an improved strength Pareto evolutionary algorithm to select only the few important genes responsible for disease identification. The method explores the whole search space for approximating the paretooptimal front that provides the optimal solution. Each chromosome in the population is evaluated using three different objective functions considering the external cluster validation index, number of genes in a sample and mutual correlation between the objects separately. The experimental results of the proposed method prove the effectiveness and acceptability of the improved strength Pareto evolutionary algorithm for the purpose of important genes selection and sample categorization.

Chapter 8 is about the medical decision support system using data-mining semicircular based angle oriented facial recognition using neutrosophic logic. By adopting two-level optimizations such as micro and macro, undesirable attitude-oriented images are deleted from the input dataset and identified the similar face from specific angle-oriented images. The experimental effects records are compared to perspective-based images of various large databases such as Yale, MIT, FERET, and College Academic for identification and found to be the efficient approach than others.

Chapter 9 discussed Weighted Differentially Coexpressed Gene Network Analysis (WDCGNA) of HIV-1 Disease for Preservation Module Prediction. The preservation patterns of differentially coexpression modules are also determined for better understanding, and eigengene network of differentially coexpression modules are built to represent the characteristic expressions of modules. The analyses have discovered the strongest preservation in Nonprogressor-Acute network and identified the significant genes in HIV-1 disease progression as well as in cancer progression.

Chapter 10 has established a building block between big data approaches with computational biology to identify/target the proteins from the available genomic datasets. This chapter is an overview of discussion of brief history of advancement of DNA sequencing and associated technologies as well as characteristics features of different sequencing platforms. Next Generation Sequencing data-mining and analysis provided the genetic information related with different functions and involvement in various pathways. Network biology approach contributed for the better understanding of functional values and pathway enrichment architecture of cancer encouraging proteins from the genetic datasets.

Chapter 11 developed a Multilayer Perceptron as the learning neural network to learn and map the exercises done by stroke patients based on the feature space of the training exercise set. The proposed method is useful for day-to-day monitoring of improvement of the patient condition measurement while performing the exercises without the need to regular visit to the doctor. The Multilayer Perceptron along with back propagation algorithm is used to compute the extent of correctness of the performed exercise to measure the improvement of the patient on a daily basis.

Chapter 12 proposed a Probabilistic Neural Network model for an empirical Study on Uddanam Chronicle Kidney Diseases (UCKD) with statistical analysis. Chronic kidney disease is a threatening state of living, can be induced due to malfunctioning of kidney or its pathology. It is critical to recognize factors that precipitate risk for CKD, even in people with typical Glomerular Filtration Rate. The proposed method is able to predict the CKD in Uddanam area with an accuracy of 100%. This study will be very useful to analysts and government to make decisions for further steps about Uddanam Chronic Kidney Diseases.

Chapter 13 developed a method for detecting enhanced brain Tumor using Fractional Wavelet Transform and Artificial Neural Network. The intention of the proposed method is for the early detection of brain tumor with high accuracy, sensitivity, and specificity by using large dataset of transverse relaxation time weighted MRI scans. The authors have extended the approach of 2D-discrete wavelet transform (DWT) to 2D-fractional wavelet transform (FrDWT) and the features are reduced using PCA. Further, they classified the brain MR images with the help of artificial neural network and achieved better performance.

Chapter 14 discussed on the performance analysis of various deep learning approaches based classification models for Smartphone Sensor based Human Activity Recognition. Human Activity Recognition (HAR) is a rapidly growing research field in the domain of computer vision where sequence of data for a specified time span is collected from the sensors like accelerometer and gyroscope present in these smart devices. HAR plays a crucial role in detecting a user interaction with environment which helps in surveillance, health care, building smart environment based on human-computer interaction. The authors have used five deep learning methods such as Convolutional Neural Network (1D-CNN), Recurrent Neural Network with Long-Short Term Memory (RNN-LSTM), CNN-LSTM, ConvLSTM, and Stacked-CNN for automatic extraction of meaningful information from raw sensor data. With the effective simulation results, the authors are confident about the methods in HAR applications.

We would like to thank all the contributors and the reviewers for their contributions and dedicated efforts for the successful completion of this book. We want to specially thank to the editorial team of Elsevier for their valuable technical support and superior efforts. We hope that the work reported in this volume will motivate for the further research and development efforts in the performance evaluation of biomedical and medical domain.

Editors

Dr. Janmenjoy Nayak

Dr. Bighnaraj Naik

Dr. Danilo Pelusi

Dr. Asit Kumar Das

Chapter 1: Application of dynamical systems based deep learning algorithms to model emergent characteristics for healthcare diagnostics

David Al-Dabass ¹ , and Lela Mirtskhulava ²       ¹ School of Science and Technology, Nottingham Trent University, Nottingham, United Kingdom      ² Iv. Javakhishvili Tbilisi State University/San Diego State University, Tbilisi, Georgia

Abstract

Bioengineering dynamical systems models of healthcare variables such as temperature, heart rate, blood pressure, etc. show output patterns that are nonlinear and time-varying. Online prediction of forthcoming danger from observed patterns can aid clinicians in advance decision making. To accurately represent these variables, reduced-order paradigms of dissimilar characteristics are associated to from the new, emergent behavior dynamics. To illustrate the concepts we start with a typical example of monitoring the brainwaves with the aid of Brainwave Computer Interface (BCI) through electroencephalography (EEG). We then propose algorithms and methodologies based on dynamical systems of deep learning to formulate models for such healthcare variables. Hybrid feedback (recurrent) nets are proposed to construct deep learning models of observed trajectory patterns of these variables. Each output pattern is used in a deep learning mining process to compute the relevant parameters of an emulated pattern—the virtual systems have multilayered feedback processes consisting of a mix of virtual agent-processes forming a pyramid of active nodes, where the output pattern at any one layer is a function of the layer below it. We use simulation to compute results to illustrate the algorithms’ functionality to overcome a multitude of difficulties within the system.

Keywords

Brainwaves; Deep learning; Dynamical systems; Monitoring; Signal processing

1. Introduction

Healthcare diagnostics is a vital area that can benefit enormously from progress in computer science, including advances made within computational intelligence in deep learning, illustrated and reported by available statistical data [1], Karam et al. [2], Mirtskhulva et al. [3], and research on brain-computer interface [4]. Brain monitoring to diagnose deficiencies in biological intelligence functions is one such area that is undergoing wide research and progress such as: EEG research [5], neurotechnology to enhance brain-computer interfaces [6], and projects reported by Mirtskhulava et al. [7,8], Yalung et al. [9], Padierna Sosa et al. [10], Michalopolous et al. [11], and Bhattacharya et al. [12]. Within advances being made in computational intelligence, feedback inference networks are emerging as an active area to represent healthcare variables to model dynamic intelligent systems. The input parameters of the healthcare variables are resolved from measurements by using a differential estimation mechanism for representing the knowledge embedded within the patient. The usage of dynamic knowledge mining processes guarantees that growth of knowledge is repeatedly traced as reported in projects by Al-Dabass et al. [13–16], Baily et al. [17], and Berndt [18] to find patterns in Time Series using Dynamic Programming. Deep Knowledge, represented by means of the input variables of the patterns of evolution of knowledge first layer, is then resolved using implementation of second layer dynamic procedures such as complexity theory by Bovet et al. [19], estimation of transfer function utilizing FFT based analyzers by Crawley [20], modeling and analysis of dynamic systems by Close et al. [21], and Dewolf et al. technique for continuous time parameter estimation [22]. In the applications of deep learning, for instance, there is a necessity to find the sources of specific style sequences. Alternative functions of deep learning comprise of cyclic trends in the values of stock market and trade value of retailing and commerce, alterations in the restoration aspects of patient, and prediction of motion uncertainties in compound engineering systems as reported by Gersch [23], Kailath [24], Kalman [25], Manila et al. [26], Man [27], Schank et al. [28] and Seveance [29]. In this chapter we develop complete mathematical derivation along with simulations and instances to demonstrate the techniques that have been included as follows:

A. Hybrid Inference Networks: A multilayer analytical pyramid is proposed to mathematically express knowledge contained in biological systems. This knowledge is fundamentally time varying and require differential models to express it and obtain its movement parameters from data in the output patterns. Where regular artificial neural nets (ANNs) are employed to carry out inference, the mathematical function between input (cause) and output (effect) is time invariant and is determinable by deduction listing all input variables (causes) then using a gradual stepwise process progressing through the network layers to compute the output values. Computing from output back to input to estimate the causes that generated the output pattern, e.g., diagnostics scenarios, the procedure needs to analyze the output pattern and progress backward stepwise layer-by-layer to calculate input values (causes).

B. Deep Learning and Knowledge Mining for Healthcare Models: The concepts described above are applied to mathematical models of bioengineering systems to cover the cases where knowledge contained with the data is not fixed but varies in time. As expected the output pattern is no longer fixed with time, and is used to drive a dynamical system to discover the embedded knowledge represented by the parameters which are now dynamic. There is deeper knowledge contained within these parameters which are discovered by computation of the input driving a dynamic process as layer-2. As before, these procedures embody a dynamic section to evaluate higher derivative in time driving static expressions for parameter evaluation.

2. Deep learning applications for brainwaves monitoring

A typical example of the application of deep learning technology occurs in medicine, where monitoring involves the observation of one or several medical parameters over time focused on a specific disease. Medical monitoring is classified by the target of interest. Here we focus on a kind of neurological monitoring dealing with the brainwave-monitoring with the aid of Brainwave Computer Interface (BCI) through Electroencephalography (EEG). Brainwave-monitoring using EEG is most popular method in neurological monitoring for monitoring and diagnosing a series of neurological diseases like Autism, Alzheimer’s, epilepsy, stroke and etc. EEGLAB is the powerful tool of MATLAB allowing us to process high-density EEG dataset through an interactive graphic user interface (GUI) through independent component analysis (ICA), time/frequency analysis (TCA), and the standard method of averaging.

2.1. Brain disorder

A global rise of brain disorders and the increasing government funding for improving healthcare systems can supplement the growth of the global market of BCI (Brain Computer Interface). The noninvasive BCI technology has accounted about 85% of revenue of the total global BCI market in 2013, and demonstrated a steady growth that eliminates the need to perform surgery and implantation of BCI devices (Fig. 1.1). The worldwide statistics showed that, an income of market of BCI was about 125.21 million US-dollars in 2018. According to a forecast the market will grow at a Component Annual Growth Rate (CAGR) of 12.43%, reaching 283 million US dollars in 2025 [1]. The neural activity of the cerebral cortex is usually retrieved and used for controlling artificial (prosthetic) limbs. A noninvasive technique is becoming popular, allowing doctors to replace the invasive method by wearable devices capable to easily measure the neural activity. After gaining the data from EEG they will be preprocessed using EEGLAB. An EEG signal measurement gives us a value of currents in dendrites as a result of excitations of synapsis through the neurons inside the cerebral cortex. The signal is not capable to measure each of the neurons but capable to measure many of them. Signals are recorded using the electrodes. Dr. Jalal Karam and Dr. Lela Mirtskhulava showed that the brain parts can be specified by the electrodes positions [2,3]. BCI cannot read the mind but can detect the frequency patterns in the brain. Our thoughts, behaviors or emotions are result of the communication between neurons in our brain. Brainwaves are generated by synchronized electrical pulses from plenty of neurons communicating with each other.

Figure 1.1 Global brain computer interface market.

2.2. Brain-computer interface (BCI)

Brain is the central control unit of the human body controlling our thoughts, speech, movements and memory. Brain is responsible to regulate the function of the human organs. Brain works efficiently and automatically when it is healthy but brain disorders can cause devastating results. We differentiate three main groups of brain computer interfaces like invasive, semiinvasive and noninvasive. Invasive technique implies to use devices for capturing the signals generated by brain where surgery is required to insert them into the brain. Semiinvasive techniques implies to insert devices into the skull not into the brain [4–6].

Brain computer interface (BCI) (Fig. 1.2) is a device enabling the users to have an interaction with computers by mean of brain activity generally measured by electroencephalography (EEG), which is a physiological method for recording the brain activities through electrodes placed on the scalp surface, and is more widely used due to its noninvasive nature. In noninvasive techniques electrodes are placed on the scalp (Fig. 1.3). EEG-based interfaces are very easy to wear and avoid surgery but have poor spatial resolution. They cannot use high frequency signals because the skull can dampens signals. After capturing EEG signals can be processed to get control signals to be readable by the computers. The processing of EEG signals is a difficult task for building a high quality BCI. The EEG recording shows the mental or physical state of the subject. If EEG shows alpha brainwaves with high amplitudes in the occipital area, it means that the subject is relaxed and the eyes are closed. If the eyes are opened, alpha-waves disappear.

Figure 1.2 BCI diagram.

Figure 1.3 Invasive and noninvasive BCI [4].

2.3. Brainwaves studies

Our brain has interconnections of billions of neurons communicating with each other through electrical currents traveling throughout our neural network acting as electrical impulses. During an activation of neurons, electrical pulses are generated forming brainwaves as a result. Here are different brainwaves according to different states of activities and thoughts. Analysis and monitoring of brainwaves play crucial roles to treat neurological malfunctions and may prevent fatality. Strokes can occur during any time, even during sleep in some cases. The time of onset of a stroke at night time is less prone to be discovered. It is vital to detect the onset of a stroke because many of them emerge due to a blood lump in a vessel inside the brain, and are suitable for treatment. Mirtskhulava et al. [7,8,30] showed that the time of onset of a stroke is crucial because limbs can be affected and therefore cause loss of motor function.

Thus, monitoring brainwaves through EEG by recording any frequency changes would enable the time of beginning of the stroke to be determined, even occurring during sleep. A dataset in BDF format was used in sleep research where brainwaves were characterized by their frequencies and amplitudes measured in Hertz. According to these frequency values, it is possible to determine the symptoms of diseases. Major research projects carried by Holewa et al. [31], Wolpaw et al. [32], Yalung et al. [9], Padierna Sosa et al. [10], Michalopolous et al. [11], Bhattacharya et al. [12], Yuan [33], Caminiti, et al. [34], Nolan et al. [35], Gurumurthy et al. [36], together with stroke dataset [37], amply illustrated that fast and slow activities of the brain are intrinsically related to high frequency and small amplitude, and low frequency and high amplitude respectively. There are five main types of brainwaves (Fig. 1.4).

2.4. EEGLAB preprocessing

EEGLAB is an interactive toolbox built in Matlab to proceed data processing obtained from EEG [2]. EEGLAB is equipped with an interactive GUI giving the possibility to process high-density brain data obtained from EEG in two ways through ICA and TFA. EEG data out of the recording device is a continuous process signal. It is like measuring a difference of potential on an oscilloscope (Fig. 1.5). To make sense of data, we need to:

(1) Extract meaningful measures from brain oscillations.

(2) Compare brain data in different conditions.

(3) Assess reliable changes due to external stimuli.

Which involves the following Preprocessing steps:

(1) Collect EEG data.

(2) Import into EEGLAB.

(3) Import event markers and channel locations.

(4) Rereference/download-sample.

(5) High pass filter (~0.5–1Hz).

(6) Examine raw data.

(7) Identify/reject bad channels.

(8) Reject large artifact time points.

(9) Run ICA and reject components.

EEGLAB supports different data formats. We used the dataset generated in the BDF file (Fig. 1.6) for a number of 16 channels at each time-point. There is only one Epoch before starting to process data. The Epoch gives a duration in seconds in the entire recording with the specific sampling rate measured in Hertz (Hz) and defined as a series of time points obtained in a second. EEG data consists of time-points representing the voltage samples measured in microvolts (μV). Sampling rate values varied from 250 to 2000   Hz. When the sampling rates are high the signal's dynamics' resolution is better but the data processing time is long (Fig. 1.7).

We filtered continuous EEG data using linear finite impulse response (FIR) filtering method in EEGLAB which involved the inverse Fourier transform. We updated dataset after applying filtering. Digital filters improved the signal-to-noise ratio or line noise at 50/60   Hz. Human EEG waves in sleep state are typically in a range of 100–500   μV in amplitude, we used a signal with a 200   μV (Fig. 1.8). The filters commonly used by EEG can be low pass and high pass for attenuating high and low frequencies respectively. For example, a high pass filter with a cut-off, 0.5   Hz frequency can be used for attenuating the components of lower frequency like 0.5   Hz and pass the components of higher frequency higher than 0.5   Hz. The aim of an ICA is to generate the available independent signals in the channel data. When the data were full rank the channels numbers and independent components (ICs) numbers are quite similar (Figs. 1.9 and 1.10). We used EEGLAB to visualize and model brain dynamics based on events and an individual dataset.

Figure 1.4 The frequency patterns in the brain.

Figure 1.5 Component activities.

Figure 1.6 BDF